A Whitepaper

Your Archive
Is Not a Memory

Why uploading twenty years of your work into an AI doesn't work — and the four disciplines that make it work.

14 pages · free · no gate on the ideas

Your problem was never finding the 2014 presentation. It was that starting the new proposal from memory is faster than the archaeology — so twenty-five years of accumulated judgment sits permanently about forty minutes out of reach.

Which in practice means it sits at a distance of infinity. So you write from the top of your head. The work is good, because you are good. It is also a fraction of what you know.

Generative AI makes a genuinely new question askable. Not where is it — that was solved fifteen years ago. But: what have I learned across my career, what evidence do I actually have for it, and where does my own material contradict me?

This paper is about why that question is harder to answer than it looks.

Everyone tries the same thing first

Point a capable model at everything. Dump the archive into a project. It produces a demo that feels remarkable for about a week — then it starts producing confident garbage, and it takes a while to work out why.

Flattening

Your 2009 closeout, your flagship framework, a think-piece you abandoned, and someone else's deck you saved for reference are four documents of equal standing to a model reading a folder. So it blends them.

Numbers without baselines

The whitepaper says 140 accounts. The workbook it was built from says 112. Nobody reconciled it in 2022 and a model will repeat whichever it finds, fluently, in your name, in front of a client.

The agreement machine

An LLM over your own corpus is a machine for agreeing with you — you wrote the corpus. The valuable question is the opposite one, and a pile of documents structurally cannot answer it.

Confidentiality

Your best proof points came from client engagements and most can't be repeated as written. You know which is which. Your file system doesn't.

What's in the paper

The full claim-and-evidence model we run in live engagements — including the linking rule that makes the two structurally incapable of drifting apart, and the provisional-numbers discipline that keeps an unreconciled figure out of client-facing work.
Why your system should not be canonical for your manuscript — and the read-only projection pattern that lets Word stay Word without ending up with two live copies, neither of them trustworthy.
A four-signal rubric for whether any of this is working, from the polite floor to the one signal the entire thesis rests on.
A worked example: four items, four formats, fifteen years, and one idea the consultant had been saying since 2009 without ever writing it down. Then the part that would ruin her afternoon.
Two boring questions to answer before you start — the ones that end more of these projects than any technical problem does.
A ninety-minute exercise you can run alone, today, that produces no software and no AI and is the precondition for both.

A number is a measurement someone took in a context, not a truth. The moment a figure gets promoted to canon it stops being checkable and starts being repeated.

Who this is for

Written for

Independent consultants and advisors, fifteen-plus years in
Boutique firm founders whose IP is the asset
Authors and practitioners with a book, a methodology, or a body of articles
Anyone who has been asked "where's that number from?" and had to go looking

Probably not for

A thin corpus — under fifteen years, or little accumulated material. The value has to come from what you capture going forward, which is a different system
Anyone wanting a tool recommendation. The paper is about the discipline, not the software
Personal knowledge management. That optimizes for what you read; this is about what you produced

Get the paper

Fourteen pages. No sales call attached, and the ninety-minute exercise at the end works whether or not you ever speak to us.

One email with the paper. Nothing else unless you ask. No list, no sequence, no unsubscribe theatre.

Or just download it directly — the ideas aren't behind the form.

Your ArchiveIs Not a Memory